Summary: Create a maintained source of approved business facts before automating customer answers. Record uncertainty and handover rules alongside services, coverage and booking information.
An AI knowledge base for tradies should begin with what the business can confirm, not what a chatbot can convincingly say. If the source material contains outdated hours, vague service promises or contradictory pricing statements, a polished answer can still be wrong.
The knowledge base is the business's maintained reference. The chatbot is one possible user of it. Staff, website writers and intake systems can benefit from the same clear facts.
Start with a small approved set
List the services offered, actual coverage, contact hours, enquiry process and important limitations. Include the approved source and the person responsible for each item.
Do not copy every page and old document into the system indiscriminately. First identify conflicts and remove outdated information from the material the assistant is meant to use.
NIST's AI Risk Management Framework provides a broader reference for managing AI use and evaluation. The worksheet below is a practical small-business approach, not a claim of certification or compliance with that framework.
Record facts and boundaries together
| Topic | Approved information | Boundary to record |
|---|---|---|
| Services | Work the business actually provides | What is excluded or assessed separately |
| Coverage | Current service area | Conditional locations need confirmation |
| Booking | How an appointment is confirmed | An enquiry is not a reservation |
| Pricing | Approved explanation of the process | No invented job-specific quote |
| Availability | Current, maintained information | Do not promise a slot from stale data |
| Technical questions | Approved general information | Hand over job-specific judgement |
Write boundaries in ordinary language. “Never guarantee attendance from this document alone” is clearer than a vague instruction to be careful.
Give uncertain answers a useful route
An assistant should be able to say that a team member needs to confirm something. That response is useful when it explains the next step rather than repeatedly apologising.
An illustrative fallback is: “Our team needs to confirm that for your job. Please share the service and suburb, and we can arrange the appropriate next conversation.”
Use only the information and contact route the business actually needs. Do not make the customer provide a complete job file simply because the chatbot is uncertain about one fact.
Write entries that can be maintained
Keep each entry focused on one topic. Include a last-reviewed date and an owner. If a fact depends on a live system, state that dependency instead of copying a value that will soon be stale.
For example, a calendar may be the source for actual availability, while the knowledge base explains the booking process. Mixing those roles can cause the assistant to repeat yesterday's opening as though it is still available.
Avoid burying exceptions in long marketing paragraphs. The person maintaining the system should be able to find the operative fact quickly.
Test questions with missing information
Try ordinary questions and awkward ones: “Can you come now?”, “What will it cost?”, “Do you work in this suburb?” and “Can you guarantee this will fix it?”
Compare each answer with the approved facts. Check whether the assistant admits missing information, uses the correct route and avoids inventing details to finish the conversation.
Also test conflicting source text deliberately in a non-public environment. The system should have a clear authority order or flag the conflict for correction. A prompt alone is not a guarantee that every answer will remain within bounds.
Review changes before customers see them
When a service or policy changes, update the source, test affected questions and remove superseded versions from active use. Keep a recoverable previous version where the tool supports it.
Assign someone to review unanswered questions and factual errors. Those records can show where the knowledge base needs a clearer entry or where a human-only process is more appropriate.
Google's helpful-content guidance also supports using verified, useful information when adapting this material into public pages. The goal is a consistent explanation of the real business across people, pages and systems—not a larger collection of confident-sounding answers.
For help connecting a clear business process with enquiry automation, see Tradie Local’s automated lead follow-up.
Sources & further reading
Frequently asked questions
Can I just upload the whole website?
You can use website material as a source, but review it for conflicts and outdated claims first. More text does not automatically mean more reliable answers.
What information should be included first?
Start with services, coverage, hours, booking steps and clear limitations. Give each topic an owner and approved source.
Should the knowledge base contain live availability?
Only if it stays reliably connected to the source of truth. Otherwise explain the process and require confirmation from the live system or team.
Can a strong prompt prevent every wrong answer?
No. Test the system, limit its actions and maintain a human handover route. Instructions alone are not a complete control.
How often should I update it?
Update affected entries whenever business facts change, and review unanswered questions or errors regularly. Assign responsibility so maintenance is not optional.
